A LANDSCAPE-AWARE, OPERATOR-ASSISTED, AND ADAPTIVE LEARNING-BASED EARLY DETECTION, VERIFICATION, AND RISK ANALYSIS SYSTEM.

Abstract

The invention relates to an AI-powered surveillance system and method used for monitoring large areas, early incident detection, risk assessment, and alarm verification. While specifically developed for the early detection and verification of forest fires, it can also be used to monitor the activities of people, vehicles, wild animals, and similar moving objects. The invention comprises a pan-tilt-zoom (PTZ) camera system (20) with high-resolution optical camera (21) and thermal sensor (22) detection capabilities, local AI processing units (10), a central monitoring, control, and recording server (70), a fixed station operator interface (30), an environmental sensor unit (80), and a data communication network (40) working together.The system automatically analyzes the area to be monitored using only the fixed station starting coordinates (103) and the digital terrain model (DEM) database (107) data; it determines terrain obstacles, line-of-sight vectors (106) and sky regions up to the maximum terrain height vertical atmospheric threshold area (elevation offset) (129) ceiling, and creates the optimum scanning plan without the need for any manual calibration. In the invention, information obtained from the high-resolution optical camera (21), thermal sensor (22) and mobile application unit (60) is evaluated with a multi-sensor data fusion approach, and the detections generated by artificial intelligence algorithms are processed not as definite alarms but as preliminary alarm candidates.Early alarm candidates are linked by cross-validation with environmental sensor data, historical event logs, risk parameters, and operator feedback, generating multivariate dynamic risk priority scores (Ps) and enabling more frequent and higher-resolution scanning of high-risk areas. The invention also enables the continuous updating of artificial intelligence models via a federated learning architecture using operator validations and field data, contextual suppression of recurring false alarms, and geographically accurate calculation of the calculated target location coordinates (104). When necessary, autonomous aerial vehicles (drones) (50) are deployed to alarm areas to perform close-range optical and thermal verification within the safe flight altitude threshold (52), and the obtained data are included in the system as a secondary confirmation layer.

Description

1 TARIFF TERRAIN-AWARE, OPERATOR-ASSISTED, AND ADAPTIVE LEARNING BASED ON EARLY DETECTION, VERIFICATION AND RISK ANALYSIS SYSTEM TECHNICAL AREA 5 This invention is important for the early detection of forest fires, among other things. Detection, human, vehicle and wildlife detection, environmental anomaly monitoring and risk analysis. Developed for these operations; fixed surveillance stations, pan-tilt-zoom (PTZ) cameras, motion sensor systems, optical and thermal sensors, computer vision, artificial intelligence intelligence and deep learning algorithms, digital terrain modeling (DEM), meteorological data analysis, 10 geographic information systems, autonomous aerial vehicles (drones), mobile device integration, and human- An integrated early warning, incident detection, verification, and decision-making system covering machine interaction areas. It relates to the support system. The invention is particularly suitable for open terrain or forests that are large, rugged, and where visibility conditions are constantly changing. The detection sensitivity of fixed surveillance systems used in these areas increases by 15 as the distance increases. drops, atmospheric events generating a high rate of false alarms, continuous live video processing High CPU / GPU load caused by the necessity of detection, blind spot detection shortcomings and contextual knowledge gained through experience by operators working in the field a data processing architecture that solves technical problems such as not being able to utilize it sufficiently, It relates to the scanning and adaptive learning method. In this context, the invention covers a surveillance area of ​​only 20 This divides the land into equal-area sectors based on physical land size, not angularly. Dynamically prioritizing sectors and using AI-based sensing results to guide operators. It offers a structure that can continuously train through feedback. STATE OF KNOWLEDGE IN TECHNOLOGY 25 Commonly used approaches in current forest fire early detection systems include fixed approaches. Angular camera scanning, continuous live video streaming, uninterrupted processing, and pre-laboratory They are based on static artificial intelligence models trained in a specific environment. However, these approaches, serious technical limitations and problems in real field and natural conditions as listed below It contains: 30  The Problem of Iso-Angle Scanning and Geometric Focusing: Fixed-angle or iso-angle scanning In traditional systems that do this, areas near the camera occupy small physical spaces. While some regions encompass a wider area, more remote regions represent much larger geographical areas. This situation, 2 Smoke, flames, and temperature changes at long distances are detected at the pixel level in very small increments. This causes it to remain there, reducing detection sensitivity and preventing the fire from spreading in the early stages. This prevents detection. However, at close range, atmospheric pressure... Fluctuations, fog, dust, or cloud movements are interpreted as false alarms by the system. is produced. In addition, traditional systems, the atmosphere above the horizon 5 distinguishing the layer or geographical obstacles (mountains, hills, etc.) outside the line of sight (LoS). because it could not, the camera angle and processing power were adjusted to avoid fire risk or physical hazards. It wastes resources in inaccessible areas.  High Processor Load and Bandwidth Cost: Uninterrupted scanning of large areas Processing high-resolution live video streams on both local processing units 10 This also creates excessive CPU and GPU load on central servers. This situation, while increasing the system's energy consumption and hardware costs, scalability and network It limits bandwidth.  Static Learning Approach and Exclusion of Operator Experience: Existing artificial intelligence Most intelligence-based systems operate using a static learning approach. 15 Models that were trained before installation are then adapted to the field using site-specific methods. from smoke behavior, seasonal variations, and local atmospheric effects cannot be nourished; incorrect, temporary or context-specific errors occurring during actual use. It cannot learn from alarms and continues to generate the same type of false alarm. However Field operators, over time, learn to distinguish region-specific anomalous behaviors. They gain a high level of experience that they can utilize. This experience and operator feedback... The inability to dynamically transfer the system's artificial intelligence model parameters results in false alarms. This is the biggest technical shortcoming in the inability to reduce these rates.  Lack of Moving Object Activities and Multilayer Validation: A significant proportion of fires are directly related to human activity. Nevertheless, 25 In current systems, movement of people or vehicles in scanning areas poses a fire risk. It is not dynamically correlated with the analysis and does not affect the screening priorities. Additionally, the presence of blind spots can affect images from mobile devices or autonomous aerial vehicles (drones). Because it is not supported by incoming secondary confirmation layers, it can be directly detected with a single sensor alert. An alarm is presented to the operator, which increases the operator's workload. 30 Due to all these technical limitations and shortcomings;  Possessing physical terrain awareness and geographical prioritization skills,  Dynamically optimizes processor and network resource usage,  Using multi-authentication mechanisms integrated with mobile and autonomous vehicles, 35  Able to continuously and adaptively learn in the field through operator feedback, 3  A novel organism capable of developing adaptive behavior specific to the changing conditions of the screening area. An integrated system architecture and data processing method are required. PURPOSE OF THE INVENTION AND DEFINITION OF THE TECHNICAL PROBLEM. The main purpose of this invention is to detect forest fires and moving objects in the early stages, at a minimum of 5 With a low false alarm rate, optimized processor load, and direct adaptation to field conditions. operator-assisted and adaptive learning capability that enables detection The goal is to create a trainable artificial intelligence system and methodology architecture. To this end... The technical problems that are targeted for resolution are as follows:  Ignoring the physical geographical reality of the surveillance area and at long distances 10 Detection imbalances resulting from scanning methods that reduce sensitivity,  The camera is unable to filter out obstacles outside the line of sight and the sky layer. and the waste of processor resources,  High hardware and bandwidth requirements caused by the need for continuous live video processing. loads, 15  High false alarm rates resulting from atmospheric events and environmental variables,  Feedback and experience from seasoned operators influence AI decisions. inability to dynamically incorporate it into the mechanism,  The inability to reflect dynamic fire risks related to human activities in the screening strategy, and Perception deficiencies in blind spots. 20 The invention addresses these aforementioned technical problems;  Physically approximate the surveillance area using Digital Terrain Model (DEM) data. dividing into coaxial sectors and radial bands of varying angular width,  Determine terrain obstacles using line of sight (LoS) analysis and then add a threshold (offset) elevation 25 by adding and excluding unnecessary fields from scanning,  Panoramic static visualization screen that does not require refreshing and only Reduces workload by providing situational awareness for the operator through live video streaming triggered by alarms. to raise awareness,  Operators receive approved / rejected alarm feedback with time, location, and meteorological data. 30 By labeling, the AI ​​model parameters are updated on the central server and locally. by distributing to units to ensure regional adaptation,  Dynamic alarms for recurring false alarm sources in geographical and temporal contexts. filtering with suppression rules, 4  Mobile device data and autonomous aerial vehicles (drones) for blind spot compensation and secondary confirmation to integrate into the system  Optical, thermal, and external meteorological sensor data are subjected to a multilayer data fusion process. By running cross-validation and suppression rules,  The system has the ability to proactively provide warnings by performing pre-event risk analysis and geographical prioritization. 5 bring, It provides technical solutions in this way. BRIEF DESCRIPTION OF THE FIGURES Figure 1: This is a block diagram showing the overall hardware and data processing architecture of the system. 10 Figure 2: Perspective showing the structural and hardware components of the fixed observation station. It is appearance. Figure 3a: Schematic illustrating the method of creating a survey area based on terrain awareness. It is an appearance and will provide approximately equivalent land area at different distance bands. It is a normalized trapezoidal segmented image frame view, 15 Figure 3b: Schematic illustrating the method of creating a survey area based on terrain awareness. It is the view and the marking of areas within the segments that are not desired to be scanned. This is a schematic view illustrating the phenomenon. Figure 4a: Schematic showing panoramic static visualization and resource optimization method. This is the view and the top 20 showing radial scan bands and normalized sectors / segments. It is appearance, Figure 4b: Schematic showing panoramic static visualization and resource optimization method. It is a view and a graphical representation showing the relationship between scanning bands and elevation in a sample land section. It is appearance, Figure 4c: Schematic showing panoramic static visualization and resource optimization method 25 It is a visual representation and modeling of Digital Terrain Model (DEM) data as a representation of the real terrain. It is the appearance, Figure 4d: Schematic showing panoramic static visualization and resource optimization method. It is a view and shows the elevation intersection relationships with scanning bands in a sample terrain. It is a perspective view. 30 Figure 5a: Schematic view showing the codomain and multiband scanning mechanism and radar. This is the visual representation of radial scan bands and normalized sectors on the interface screen. Figure 5b: Schematic view showing the codomain and multiband scanning mechanism, and is oblique. It is a geometric representation of the scanned areas calculated in the plane. Figure 6: Flowchart illustrating the AI-powered multi-sensing and decision support mechanism. 35 It is a diagram. Figure 7a: Operator interface, feedback confirmation, adaptive learning, and trainability steps. This is a schematic view showing the operator interface screen and the confirmation mechanism view. Figure 7b: Operator interface, feedback confirmation, adaptive learning, and trainability steps. This is a schematic view showing the alarm confirmation and feedback screen view. Figure 7c: Operator interface, feedback confirmation, adaptive learning and trainability steps 5 This is a schematic view showing the alarm termination confirmation and event outcome screen. Figure 8: Flowchart illustrating the algorithm for integrating meteorological and soil data. Figure 9: Interface showing the cross-validation screen and its components with a mobile data source. It is the appearance. Figure 10: Showing the autonomous aerial vehicle (drone) based secondary physical verification system. This is a schematic view. Figure 11: Analysis of unauthorized movement of people and vehicles as an indicator of fire risk. It is a flowchart showing the process. Figure 12: Flowchart showing the central server's operating algorithm. Figure 13: Flowchart illustrating the local data processing algorithm of meteorological and soil sensors. 15 It is a diagram. REFERENCE NUMBERS OF PARTS / COMPONENTS 10. Local AI processing unit 11. Data communication unit 20 12. Power supply pole 13. Combined thermal and optical sensor 20. Pan-tilt-zoom (PTZ) camera setup 21. High-resolution optical camera 22. Thermal sensor 25 23. Pan-tilt mechanism 24. Reference sensors (Position, orientation, and tilt) 30. Fixed station operator interface 31. Fire alarm coordinate indicator 32. Alarm start timestamp 30 33. Fire verification selection area 34. False alarm selection area 35. Operator authentication field 36. Fire termination timestamp 37. Fire termination entrance area 35 38. Event outcome explanation area 6 39. Visual alarm icon 40. Data communication network 41. Unauthorized person or vehicle 50. Autonomous aerial vehicle (Drone) 51. Threshold autonomous flight route 5 52. Safe flight altitude threshold (Offset) 53. Digital terrain model (DEM) vertical cross-section. 54. Autonomous aerial vehicle starting position 55. Target alarm point 56. The shortest straight-line distance vector is 10. 60. Mobile application unit 61. Mobile target alert point 62. Mobile device location and altitude data 63. Mobile device orientation angles (Pan-tilt) 64. Mobile device atmospheric data sensor 15 65. Mobile data transmission control 70. Central monitoring, control and recording server 80. Environmental sensor unit (Meteorology and soil) 81. Solar and atmospheric radiation sensors 82. Soil moisture and characteristic sensors 20 83. External atmospheric data 84. Real-time measurement data 85. Meteorological critical values 86. Soil characteristic data 87. Critical threshold decision logic layer 25 90. Central monitoring, recording and control interface 100. Maximum scannable area outer limit. 101. Circular scanning radii (R1, R2...Rn) 102. Co-domain scanning sectors (S1, S2...Sn) 103. Fixed station starting coordinates (X0, Y0, Z0) 30 104. Calculated target location coordinates (X, Y, Z) 105. Two-dimensional alarm coordinates (X, Y) 106. Line of Sight Vector 107. Digital Terrain Model (DEM) database 108. Vertical field of view (Tilt FOV) 35 109. Horizontal field of view angle (Pan FOV) 110. Physical land area corresponding to the first sector. 7 111. Physical land area corresponding to the second sector. 112. Physical land area corresponding to the third sector. 113. Physical land area corresponding to the fourth sector. 120. Static panoramic screen 121. Normalized scanning sector screen components 5 122. Passive screen areas excluded from scanning. 123. Excluded atmospheric region screen area 124. Counterclockwise boundary horizontal angle (-PAN) 125. Clockwise limit horizontal angle (+PAN) 126. Lower vertical limit angle (-TILT) 10 127. Upper vertical limit angle (+TILT) 128. Masked land area screen space 129. Maximum terrain height vertical atmospheric threshold area (Elevation offset) 131. Live optical viewing screen 132. Live thermal monitoring screen 15 133. Scanning radar screen 134. Scanning radar screen center 135. Live thermal display target indicator. 136. Live optical display target indicator 137. Panoramic screen real-time scanned band display 20 138. Panoramic screen horizontal-vertical angle synchronization. 139. Radar screen geographic north reference mark. 140. Radar display control and menu area DETAILED DESCRIPTION OF THE INVENTION 25 The invention involves area optimization based on digital terrain modeling (DEM), and multi-sensor data fusion. contextual meteorological assessment, operator-assisted adaptive learning, risk index generation. and by offering a holistic system that integrates distributed computing architecture, it excels in the technical field. It differs technically from traditional fire detection systems. This system... It is not just a system that detects fires instantly; it also identifies increased risk in advance and prevents false alarms. minimizing rates, increasing hardware resource efficiency, and providing a scalable early warning system. and provides risk analysis infrastructure. 1. General Structure and System Architecture of the Invention The system in question is the subject of the invention, which exchanges data seamlessly with each other via a data communication network (40). distributed 35 consisting of multiple hardware and software components that perform and coordinate operations 8 It has a process architecture. This system architecture takes place before the occurrence of fire risk. able to assess environmental conditions, detect fire outbreaks in the early stages, and By verifying the detected events with operator support, its own artificial intelligence model develops over time. It is structured in a way that allows for the improvement of its parameters. The system, whose general block diagram is shown in Figure 1, includes at least one fixed surveillance station, these 5 at least one local artificial intelligence processing unit (10) that is integrated into or connected to the station The data at the station is monitored locally and the operator provides feedback inputs on the fixed system. The station operator interface (30) is where all field data is collected, processed and managed. a central monitoring, control and logging server (70) and central level management It includes the central monitoring, recording and control interface (90). The system's alternative or preferred 10 in a configuration that includes an optional environmental sensor collecting external data from the field. unit (80), at least one optional autonomous aerial vehicle (drone) providing secondary physical verification (50) and at least one optional mobile application that provides mobile data streaming with blind spot compensation. Unit (60) is included in the architecture. In this architecture, each fixed observation station has its own a 15 capable of independently generating local decisions through a local artificial intelligence processing unit (10). It is structured as follows: However, these stations are connected to a central monitoring, control, and recording server. (70) by synchronizing with each other and with central operators throughout the system It creates a common federated learning and contextual decision-making mechanism. 2. Structure and Function of the Fixed Observation Station Figure 2 shows the detailed structural architecture of the fixed observation station, fire and object 20 It constitutes the physical data collection layer of the sensing process. The station in question... structurally; it is located on an energy supply pole (12) and has high high-resolution optical camera (day vision camera) (21) and thermal sensor (22) a pan-tilt-zoom (PTZ) camera setup (20), pan-tilt mechanism (23), reference sensors (position, orientation and tilt) (24), a data communication unit (11) and local artificial intelligence processing 25 It contains unit (10). Optionally, station according to system requirements. A combined thermal and optical sensor (13) can also be positioned on it. The high-resolution optical camera (21), which also functions as a day vision camera, is visible by working in the spectrum, visual anomaly cues related to smoke and flame formation. The thermal sensor (22) detects infrared independently of the visible spectrum. By monitoring temperature distributions on the belt, a sudden temperature increase is observed even though the flame phase has not yet begun. This makes it possible to detect potential fire sources at an early stage. The aforementioned high-resolution optical camera (21) and thermal sensor (22) enable local artificial intelligence. to form a complementary data fusion layer under the control of the processing unit (10) He / She is working. 35 9 Pan-tilt mechanism (23), vertical field of view angle of the cameras (tilt FOV) (108) and horizontal view rotation of the field of view (pan FOV) (109) axes and optical zoom This enables the pan-tilt mechanism (23) to be used in traditional systems. It does not perform a continuous random or fixed equiangular scanning cycle as is the case. Scanning movements are performed by the local artificial intelligence processing unit (10) on the digital terrain model (DEM 5 According to the co-domain scanning sectors (102) calculated based on the data of the database (107). Thus, the motor movements of the pan-tilt mechanism (23) are optimized and executed. mechanically optimized, and the system's hardware lifespan and sensing efficiency are improved. is being increased. Reference sensors (position, orientation and tilt sensors) (24); 10 degrees relative to the station's geographic north direction, yaw, pitch, and roll slope references and position by instantly determining the real-time image pixels from the cameras, the actual geographical information of the earth. This ensures accurate matching with the coordinates. This orientation data is particularly important for digital terrain. Three-dimensional line-of-sight vector (106) and blind model (DEM) performed with database (107). It is used as input in point calculations. 15 3. The Role and Integration Capabilities of the Local AI Processing Unit The local AI processing unit (10) processes all raw visual and data collected at the fixed monitoring station. The main tool that performs the initial evaluation of thermal data using edge computing methods. It is a component of the local artificial intelligence processing unit (10); high-resolution optical camera (21) and thermal 20 snapshots from the sensor (22), from the external environmental sensor unit (80) It analyzes contextual meteorological data. Environmental sensor unit (80), It is controlled by the local artificial intelligence processing unit (10) at the station. with solar and atmospheric radiation sensors (81) within the environmental sensor unit (80) instantaneous measurement data (84) collected through soil moisture and characteristic sensors (82) 25 Time intervals and comparison criteria are managed with this local control architecture. Instantaneous measurement data (84), predefined meteorological critical values ​​(85) or soil If the characteristic data (86) exceeds its critical limits, the local AI processing unit (10) The critical threshold decision logic layer (87) within it is triggered and the protective warnings generated Central monitoring via fixed station operator interface (30) and data communication network (40), 30 All these contextual and generated information are transmitted simultaneously to the recording and control interface (90). Time-dependent raw / processed data on the central monitoring, control and recording server (70) They are recorded relationally. Also autonomous aerial vehicle (drone) (50), fixed station operator interface (30) or central It can be remotely directed and managed via the monitoring, recording and control interface (90). Local 35 artificial intelligence processing unit (10), fire alarm calculated when an alarm condition occurs Based on the coordinate indicator (31) and from the digital terrain model (DEM) database (107) By processing the data it receives, it determines the threshold autonomous flight path (51) for the autonomous aerial vehicle (drone) (50) and The autonomous aerial vehicle automatically calculates the safe flight altitude threshold (52). (drone) (50), depending on its distance to the alarm point and battery / range limits, local 5 It can communicate directly with the artificial intelligence processing unit (10); mobile networks, radio frequencies or a flexibility that can be dynamically controlled via satellite communication channels It has. The local AI processing unit (10) is not just a "data transmitter" but a continuously learning one. It is a decision component that can make contextual evaluations. Local AI processing unit (10), transmitting detection results raw to the central monitoring, control and recording server (70) 10 instead, by pre-evaluating with local deep learning models, only specific When anomalies and thresholds are exceeded, it generates "alarm candidates". These alarm candidates... It does not directly constitute a definitive alarm status. False alarms are eliminated through software. supporting legal verification processes, realistically assessing environmental conditions In order to interpret and optimize the operator's operational workload, the operator, 15 It has been incorporated into the system as a conscious part of the decision-making process. At this stage, the system... Fixed station operator interface (30) or central monitoring, recording and control interface (90) It works by consciously triggering human-machine interaction. 4. Operator Interface, Interaction, and Dynamic Adaptive Learning Cycle The fixed station operator interface (30), the details of which are shown in Figure 7a, Figure 7b and Figure 7c, 20 data received from the operator's fixed monitoring station via the data communication network (40) It monitors, evaluates the generated alarm candidates, and provides structured feedback to the system. It provides a human-machine interaction layer. These verification and feedback processes, It can be run locally via the fixed station operator interface (30), as well as centrally. remotely and centrally via the monitoring, recording and control interface (90) 25 This can be accomplished by detecting an anomaly by the local artificial intelligence processing unit (10). When the alarm candidate is generated, the alarm candidate is displayed on the fixed station operator interface (30) or is submitted for operator approval via the central monitoring, recording and control interface (90). The operator reviewed the relevant alarm candidate and determined that the event was a genuine fire. Confirming via the verification selection area (33) or a permanent / temporary false alarm source 30 (factory chimneys that smoke at certain intervals, reflections from solar power plants, hot surfaces) etc.) by marking the feedback loop via the false alarm selection area (34). It initiates permanent or temporary false alarms generated at fixed surveillance stations. approvals are primarily incorporated into the installation and calibration parameters of the relevant station. recorded locally; simultaneously central monitoring, control and recording server (70) 35 It is backed up relationally with the relevant station ID. 11 Confirmed feedback actions performed by the operator are only for the immediate alarm candidate. It is not used for the purpose of closure or approval. The decision is made by the operator; co-domain scanning sectors where the alarm occurred (102), alarm start timestamp (32), instantaneous external atmospheric data (83) and detection type (flame / smoke / object) parameters labeled and sent as a relational data set to the central monitoring, control and recording server (70) 5 is transmitted. The central monitoring, control and recording server (70) receives these collected labeled data sets. using the local artificial intelligence processing unit (10) to develop deep learning models and systems retraining and optimizing the decision-making mechanism parameters across the board It provides. Persistent false alarm sources, based on operator confirmation, are geographically located and time-bound. By being labeled with a stamp, the system applies geographical and temporal alarm suppression rules. The system is transforming these areas into contextual ones so that they will not generate potential alarm candidates in the future. It records this as a filter. Thanks to this adaptive learning mechanism, the system; the operator's site-specific geographical experience, environmental changes that have occurred over time, Seasonal and atmospheric variations can be attributed to an external manual data collection process or expert 15 without requiring intervention, natural usage process during normal operation It learns within the system and automatically adapts to local field conditions. This is a natural adaptation. As a result of this process, the system has dynamically learned local sources of false alarms and identified risky ones. This would identify the sectors and integrate operator experience into a mathematical model. In this way, false alarm detection rates are minimized, operator intervention and 20 Operational workload is reduced, and the predictability and decision reliability of the system are increased. Fire termination processes include fire termination timestamps (36), fire recorded via the termination entry field (37) and event outcome description field (38) The system's event history is being added to the database. 5. Centralized Server and Distributed Learning Structure 25 Central monitoring, control and recording server (70); data communication network within it (40) from all fixed surveillance stations, mobile application units (60) and instantaneous measurement data (84) and operator feedback from environmental sensor units (80) The central monitoring, control and recording server (70) collects the collected data. By performing long-term statistical and relational analyses, we identified 30 common factors across the system. producing federated learning models and updating artificial intelligence model parameters. This distributed processing is distributed to local AI processing units (10) in the field. Thanks to its architecture, operational experience and alarm gained in a specific surveillance area. Verification data, in a controlled and context-sensitive manner, monitors other surveillance on the system network. This is reflected in the models of the stations; however, each station has its own local micro-35 12 We continue to maintain parameter configurations specialized according to environmental conditions. is doing. 6. Panoramic Static Visualization and Graphic Resource Optimization Unlike traditional surveillance systems, the operator continuously monitors the live video stream. 5 perceptual fatigue, distraction, and high-tech equipment resulting from the necessity The invention is based on graphics resource optimization to eliminate processor load. A static visualization method is presented. As shown in Figure 7a, the fixed station operator interface (30); static panoramic display (120), scanning radar display (133), live optics It consists of the main blocks of the monitoring screen (131) and the live thermal monitoring screen (132). Scan 10 radar screen (133), circular scan showing the scanning area in bird's-eye view geometry. scanning radar screen marking radii (101) and co-area scanning sectors (102) Orientation via the central (134) and radar screen geographic north reference marker (139) It has a circular structure. Thermal, optical and radar screens are used during the scanning operation. It operates simultaneously and synchronously under the control of the local artificial intelligence processing unit (10). 15 Static panoramic screen (120), initial setup, calibration or of fixed observation station. During the periodic renewal phase, the maximum scannable area can be used by using the pan-tilt mechanism (23). Reference obtained by scanning the surroundings 360 degrees within the outer boundary of the area (100) It is created by stitching together images. This is the static panoramic screen in question. (120) not a continuously streaming live video, but a geometric, geographical and visual 20 of the scanning area It is a static background layer representing the reference. On the static panoramic display (120), vertically lower vertical limit angle (-TILT) (126) and upper vertical limit angle (+TILT) (127) and horizontally clockwise Clockwise limit horizontal angle (-PAN) (124) and clockwise limit horizontal angle (+PAN) (125) Horizontal-vertical angle synchronization of panoramic screen using parameters (138) This is being done. To maximize operator perception of the interface, it is designed for both summer and winter and 25... It has night / day modes. On the static panoramic display (120), the digital terrain model (DEM) database (107) or in line with the analyses performed with an alternative digital elevation model database; Scanning that grows from near to far and is normalized based on actual land size. The sector screen parts (121) are defined. On this pixel mapping layer, view 30 According to the line of sight vector (106) analysis, those who are behind terrain obstacles or fire Passive areas such as buildings that do not pose a risk are considered passive screen areas excluded from scanning (122). It is being masked. In addition, it is added on top of the highest geographic elevation that can be scanned. The maximum land height is calculated by calculating the vertical atmospheric threshold area (elevation offset) (129), The upper atmospheric region that does not correspond to the land surface, the atmospheric region excluded from scanning 35 13 The screen area (123) is completely cropped from the system scan plan. Outside of scanning left passive screen areas (122) and masked land segment screen areas (128) This increases the system's scanning frequency and minimizes hardware / software processing load. is being done. To enhance the operator's spatial awareness, 5 real-time images are displayed on the static panoramic screen. The area being scanned is shown on the panoramic screen with the instantaneous scanned band display (137). It is emphasized that when an alarm candidate is formed, it corresponds to the two-dimensional alarm coordinate (105). In the incoming position, a flashing visual alarm symbol (39) or representative icon is lit to alert the operator. Live video streams (Live optical monitoring screen (131) and Live thermal monitoring are offered. screen (132)) is not operated continuously, only the local artificial intelligence processing unit (10) 10 When a real-time scan is performed by the company in the relevant sector, an anomaly / alarm candidate is detected. when or live thermal display target indicator on the operator's live monitoring screens (135) and request a manual inspection via the live optical display target indicator (136) It is triggered in this situation. Thanks to this resource optimization method, local artificial intelligence Processing units (10) CPU and GPU resources and data communication network (40) bandwidth 15 energy consumption is optimized, the operating life of hardware components is extended, and the system's energy efficiency is improved. consumption is being reduced and the rates of human-induced oversight errors in long-term monitoring are decreasing. is being reduced. 7. Establishment of a Land Awareness-Based Survey Area In traditional scanning setups, pan-tilt-zoom (PTZ) cameras typically use fixed, unipolar 20-degree cameras. It operates using step-by-step or time-based linear scanning cycles. In this approach... The cameras scan the environment at specific, fixed angular increments. However, they are not angularly equal to each other. These scanning steps are due to the slope of the earth's topography and its distance from the sensor. They do not correspond to equal physical areas on the actual land surface. As shown in detail in Figure 4, a small angular movement of the pan-tilt mechanism (23) 25 While the camera covers a narrow physical area in geographically close regions; the same angular step, at a distance a much wider physical land area due to pixel resolution loss over distances This corresponds to the geometric imbalance that causes fire signs (smoke, etc.) to be detected at long distances. (microscopic flashes of flame) reduce the likelihood of early detection, while close At longer distances, atmospheric movements or vegetation fluctuations can unnecessarily cause 30 This leads to high sensitivity and the generation of false alarms. The technical problem in the scanning method that is the subject of the invention is based on a digital terrain model. It is solved by a codomain scanning approach. The local artificial intelligence processing unit (10); numerical land model (DEM) database (107) data, fixed station starting coordinates (103), Orientation data from reference sensors (24) and pan-tilt-zoom (PTZ) camera 35 14 by processing the instantaneous optical zoom / focus parameters of the system (20) together, the maximum scannable It subjects the surveillance area within the outer boundary (100) to geometric analysis. As a result of this analysis, the scanning area is perpendicular to the pan-tilt-zoom (PTZ) camera setup (20). They have angular widths that vary depending on the distance, but are not physically present on the Earth's surface. Circular scanning radii (101) (R1–Rn) covering areas of approximately equal size and equal 5 The area is divided into scanning sectors (102) (S1–Sn) (Figure 4a and Figure 5a). Digital terrain Three-dimensional line of sight model (DEM) performed using database (107) data. Through vector (106) analysis, the scan plan is physically co-domain and optically accessible It is divided into segments. The line-of-sight analysis in question is performed using the following vector equation (Equation 1) It is defined by: 10 𝐿 = (𝑋 − 𝑋 , 𝑌 − 𝑌 , 𝑍 − 𝑍 ) Here, the vector L is the line of sight vector (106), (X_0, Y_0, Z_0) is the fixed station starting point. (103) represents the coordinates of the calculated target position, (X, Y, Z) is (104) Thanks to this vector structure, each codomain scanning sector (102) provides local artificial intelligence 15 becoming independent observation units with equal statistical significance in terms of processing unit (10). This is brought about. Thus, extreme sensitivity to local atmospheric movements at close distances. This prevents false alarms from occurring, while blind spots are minimized and distance Small plumes of smoke or thermal anomalies at a distance are pixel equivalent to events at a close distance. It is perceived with sensitivity. 20 8. Co-domain and Multi-band Scanning Mechanism and Algorithm During the initial installation phase of the system in the field, the pan-tilt-zoom (PTZ) camera setup (20) was only Reference orientation parameters include the geographic north direction and the horizontal zero axis. It is mounted by taking the starting coordinates of the fixed station (103) and camera 25 into the system. After the optical parameters are entered, the local artificial intelligence processing unit (10) can process any manual input. The following operational steps are performed automatically without requiring operator intervention. It is:  Digital terrain model (DEM) database (107) data of the relevant surveillance area is obtained is done, 30  Surveillance area within the outer limit of the maximum scannable area (100), circular scan Automatic DEM-based circular scanning bands in the direction of radii (101) is leaving,  Within each scanning band, physically approximately equivalent areas on the earth's topography. screening sectors (102) are being created,  For each sector created, the pan angle and tilt angle of the pan-tilt mechanism (23) and the pan-tilt- The zoom parameters of the zoom (PTZ) camera setup (20) are calculated automatically,  By identifying the highest geographic terrain points on the digital terrain model, these elevations are divided into 5 a predefined maximum terrain height vertical atmospheric threshold zone (elevation) offset) (129) is integrated,  Regions of the upper atmosphere or sky that do not pose any fire risk, radial bands. Digital elevation data and optical line of sight vector (106) within it are taken into consideration for visibility. They are excluded from the screening plan by being included. 10 As shown in Figure 5a and Figure 5b, between each circular scanning radius (101) Even if the radial distance is determined to be linearly equal, the circular distance expands depending on the distance. In order to balance the perimeter lengths, the system corresponds to the physical land area of ​​the first sector. area (110), physical land area corresponding to the second sector (111), corresponding to the third sector physical land area (112) and physical land area corresponding to the fourth sector (113) 15 their geometries remain approximately equal to each other (Ai ≈ Aj) on the Earth's plane This normalization process is provided by a pan-tilt mechanism for each sector. (23) instantaneous vertical field of view (Tilt FOV) (108) and instantaneous horizontal view which optimize movement The area angle (Pan FOV) (109) calculations are done using the following mathematical approach (Equation 2) It is carried out in line with: 20 A ≈ A Δ𝜃 = 2A r − r , Δ𝜙 = 2A ℎ − ℎ Here; 25  Δ𝜃 : horizontal scanning angle, instantaneous horizontal field of view angle (Pan FOV) (109),  Δ𝜙 : vertical scanning angle, instantaneous vertical field of view angle (Tilt FOV) (108),  A: physical sector area, the target physical sector area size on the earth's surface,  r, r: radial band distances, radial band distance limits of the respective band,  ℎ , ℎ : elevation bands, where ℎ is the topographic elevation boundary of the respective band 30 16 It expresses. Thanks to these algorithmic calculations, each codomain scanning sector (102) is geographically The area is being surveyed in a way that is fully compatible with the topography and prioritizes risk. This is carried out at this stage. three-dimensional line-of-sight vector (106) and inter-sector overlap analyses In this way, blind spots that may occur in the field are completely minimized. Any 5 In the scanning cycle, the pan-tilt-zoom (PTZ) camera setup (20) provides a specific co-domain scan. When focusing on the sector (102), deep into the snapshot taken from the camera sensor Learning analysis is being initiated. As shown on the static panoramic screen (120) in Figure 3a; camera view perspective distortion caused by (nearby geographical areas appearing wide on the screen 10 (pixel space occupancy, compensating for distant geographic areas being compressed into a narrow pixel area) For this purpose, each scan sector is divided into pixel sub-areas called "segments". It is departing. From the bottom center point of the sensor screen (near range reference) upwards (far range reference) (distance direction) and as one moves left or right, the pixel segment to be analyzed The geometric widths of the areas are inversely proportional to the projection distortion created by the viewing angle. 15 It is reduced proportionally incrementally. This is pixel area normalization. Thanks to the layer, each of the normalized scanning sector screen parts (121) on the screen The sub-segments are scaled to represent real physical equivalent areas on Earth. and the artificial intelligence model works with homogeneous pixel precision across the entire industry. This is possible. In short, the aforementioned holistic algorithmic structure primarily involves mapping the terrain to a distance of 20 connected co-field scanning sectors (102), then each sector within itself with the earth plane It divides the system into scaled equivalent analysis segments. This mechanism prevents any external interference with the system. Suitable for different geographical features and rough terrain without the need for operator calibration. It enables the site to automatically adapt to its topographies. 9. Dynamic Determination of Screening Priorities Method 25 The system described in the invention covers the fixed surveillance station's area using only cameras. Not based on point of view; but on the physical topography of the earth that carries a fire risk factor. It analyzes according to geographical areas. Digital terrain model (DEM) database (107) data Through geometric analyses performed using; the sky layer, the distant atmosphere Scanning land areas that do not contain layers or vegetation and have no fire risk. 30 They are completely excluded from the plan. These areas are perceived from the outset. Because it was not included in the process, errors caused by atmospheric movements near the horizon line Alarm generation is prevented; active co-domain scanning sectors (102) that need to be scanned By reducing the number of scans, the overall scanning period speed of the system is increased, thus saving hardware resources. Consumption is being optimized. 35 17 Co-field scanning sectors (102) created based on the Digital Terrain Model (DEM) are linear. It is not scanned with the same frequency in a cycle. The system scans each codomain sector (102) It calculates a dynamic risk priority score (Ps). This risk priority score (Ps) is: Instantaneous temperature, relative humidity and wind speed / direction collected via environmental sensor unit (80) instantaneous measurement data (84), such as are located in the central monitoring, control and recording server (70) database. Area history alarm / fire records, unauthorized persons or vehicles detected in the scanning area (41) or data relating to moving object types, previously provided by the operator for specific sectors. Approval / rejection feedback and real-time data manually entered by the operator via the interface. The screening frequency is dynamically calculated based on priority requests. Local AI processing unit (10), critical peer 10 with high calculated risk priority score (Ps). area scanning sectors (102) with shorter time intervals and higher optical When scanning in detail with high resolution, sectors with low risk scores are identified at less frequent intervals. It includes it in the screening plan. Through this dynamic prioritization algorithm, the fire is detected. While the early detection period is shortening, unnecessary alarms may arise from low-risk areas. Their production is limited, and the system's mechanical and computational (processor) resources can be used simultaneously 15 The mechanical wear rates of the pan-tilt mechanism (23) are optimized accordingly. is being reduced. In this dynamic prioritization layer, the operator, static panoramic normalized scanning sector on the screen (120) on the screen parts (121), each Different color codes represent the real-time calculated risk status and priority levels of a sector. Tracking via graphical symbols or density changes (heat map visualization) 20 It is able to do so. The contextual and geographical feedback provided by the operator through the interface, the contiguous field survey approach fully adapted to the changing local field and vegetation conditions over time It ensures that it becomes compatible. 10. AI-Powered Multi-Sensing and Decision Support Mechanism The ability to reliably detect forest fires in their early stages requires a single 25 It cannot be performed based on physical anomaly input. Smoke, flame or sudden When indicators such as thermal increases are processed individually, atmospheric reflections and fog waves or generating a high rate of false alarms (false positives) due to controlled environmental factors. It involves risk. In the system that is the subject of the invention, fire detection is achieved through multiple detection and sensor systems. It is carried out using a fusion approach. 30 As shown in the functional flowchart in Figure 6, the local artificial intelligence processing unit (10); Deep learning-based imaging for detecting smoke anomalies in the visible spectrum. the processing module uses color and motion-based technology to detect flame characteristics at the pixel level. The temperature anomaly analysis module works with radiometric data received from the thermal sensor (22). 18 detection module and persistent or historical data recorded for the targeted sector. It runs temporary false alarm data logs simultaneously and synchronously. The local AI processing unit (10) uses the technical data from these aforementioned sensing modules. The results are not viewed as independent, singular outcomes, but as a multivariate mathematical fusion. It evaluates them together by subjecting them to its function. The comprehensive alarm decision in question is 5 The analysis is carried out in accordance with the following relational functional equation (Equation 3): 𝐴𝑙𝑎𝑟𝑚 = 𝑓(𝐷, 𝐹, 𝑇, 𝑅) Here, (D) is the optical smoke detection probability value, (F) is the color / motion-based flame detection value, (T) is the temperature anomaly value measured by means of the thermal sensor (22) and (R) is the relevant sector It refers to a matrix of past meteorological and environmental risk factors. For example, only 10 When a thermal anomaly is detected, the system does not immediately generate a definite alarm; anomalies with persistent environmental and topographic behaviors previously observed in the relevant sector (check if factors such as solar panel reflection, rock structure heating, etc.) are compatible. Similarly, when a pixelated, smoke-like movement is detected, external atmospheric signals are also considered. data (83) processed to determine whether this is an atmospheric effect such as fog or dust cloud or a real 15 Whether it is a combustion reaction is contextually confirmed. As shown in Figure 6, The numbers shown inside the circles represent the process steps (algorithm steps). This is independent of the hardware numbers in the part reference list. 11. The Concept of Alarm Candidate and Contextual Preliminary Assessment The method described in the invention aims to improve the accuracy of operational processes and legal / technical 20 To ensure validity, the concepts of "alarm" and "alarm candidate" are algorithmically distinguished from each other. It is separated. The local AI processing unit (10) processes the fusion data from the sensing modules. Instead of generating a definitive alarm when predefined critical threshold values ​​are exceeded, prioritize This creates an "alarm candidate". An alarm candidate is an indicator that represents a potential fire risk. This is the preliminary technical due diligence phase. 25 The mathematical generation of the alarm candidate and the contextual preliminary assessment process, local artificial intelligence processing unit (10) through the following functional modeling (Equation 4) It is performed by: Alarm Candidate = 𝑓(𝑉𝐷𝐴, 𝐹𝐶𝐴, 𝑇𝐴𝐴, 𝑃𝐻𝐴, 𝐷𝐸𝑀 − 𝐴𝑂) 30 Here; (VDA) video smoke analysis output, (FCA) flash / flame characteristic analysis value, (TAA) thermal anomaly analysis input, (PHA) pixel-based moving object analysis data and (DEM- AO) represents the area optimization masking parameters based on the digital terrain model. 19 The system aims to eliminate the possibility of false alarms caused by a single sensor. Analyzing different sensor data from the same event area simultaneously or sequentially (in sequence). It performs a multidimensional evaluation by comparing a single signal from a single module. It does not allow for direct alarm candidate generation; the alarm candidate is generated by cross-referencing multiple sensor outputs. It is generated by processing the verification and the dynamic risk priority score (Ps) together. 5 The alarm candidate generation mechanism in question is based on the following basic algorithmic rules: It is working:  A single anomaly signal from a single software module does not directly generate a system alarm.  Alarm candidate, cross-validation of multiple sensor outputs and dynamic priority score It is triggered if the threshold values ​​are exceeded. 10  All generated alarm candidates are presented to the operator for approval with a hierarchical context set priority. and is only activated to a definitive alarm status with operator feedback. During the generation of the alarm candidate; the co-domain scanning sector where the detection takes place. (102) identity, type of anomaly (smoke, flame, thermal), temporal duration of detection and repetition frequency and instantaneous measurement data (84) received from the environmental sensor unit (80) simultaneously 15 By bringing these together, a technical context set is created for the potential alarm. context set, which visual on the fixed station operator interface (30) of the alarm candidate It first specifies that it will be submitted for operator approval. The system calculates specific target location coordinates based on operator feedback. (104) or dynamic alarm associated with location information and time interval for screen segments 20 It is structured in a way that can create suppression rules. Within this scope, a specific segment... date ranges, days of the week, specific time slots within the day and / or predefined rules Temporary or permanent alarm assessment rules related to duration parameters It can be defined. In accordance with the defined rules, the alarm generated from the region in question. Candidates are subjected to contextual filtering, and under certain conditions, false alarms are generated by system 25. It is automatically suppressed by. Temporarily defined suppression rules, at the end of the predetermined period, automatically by the local artificial intelligence processing unit (10) It is terminated as follows. 12. Determining Fire Coordinates and Trainability through Operator Feedback 30 Algorithm The alarm candidate generated as shown in Figure 7b is the fixed station operator interface (30) and central monitoring, recording and control interface (90) via data communication network (40) It is presented to the operator for approval simultaneously. The operator, on the static panoramic screen (120) the relevant two-dimensional alarm coordinate (105), the location’s past alarm history and instant detection It can analyze the data in an integrated manner. The operator can pan-tilt if deemed necessary. by taking the mechanism (23) into manual control via the combined thermal and optical sensor (13) It can perform detailed analysis by capturing live images from the relevant area. The operator receives the following three structured feedback options via the interface for the alarm candidate being examined. He chooses one of the actions:  Verified Alarm: It is confirmed that the event is a genuine fire incident.  Temporary False Alarm: Short-term atmospheric fluctuations or controlled environmental factors. It is indicated that it is caused by factors such as authorized stubble burning, picnic smoke, etc.  Persistent False Alarm: A structural, geographical, or continuous characteristic of the relevant sector (active factory 10 systematically misperception due to (chimney, glare-producing roof surfaces, etc.) It is stated that it was produced. Temporary false alarm and permanent false alarm data generated in line with these elections are artificial. It is processed and evaluated by the intelligence model using different algorithmic weights. Alarm confirmation 15 mechanism; anomaly detection by local artificial intelligence processing unit (10) -> alarm candidate Creation -> Operator's selection via interface -> Automatic video / metadata recording -> Updating temporary / permanent alarm suppression rules -> returning tagged data to the dataset. Its feeding schedule operates in chronological order. Marked as a persistent false alarm. By updating local and central AI model parameters for geographical areas, the future will be 20 It automatically prevents the generation of alarm candidates at the same coordinates and under the same threshold conditions. Temporary alarm confirmation data shows that human-caused smoke is particularly prevalent during specific time periods. It is used in activities that reduce false alarm rates without blinding the system. All these learning and local adaptation processes take place during the normal operational flow of the system, 25 within the natural usage cycle without the need for an external manual data labeling process is being carried out. 13. Digital Terrain Model (DEM) Based Three-Dimensional Position Estimation Method In traditional surveillance systems, fire detection is mostly achieved through an optical warning or a camera. It is limited only by the perspective of the viewpoint. This linear data constraint will determine the intervention to be sent to the field. This leads to additional reconnaissance activities by the teams to verify the location and intervene. It extends the duration. In the system that is the subject of the invention, the local artificial intelligence processing unit (10) When an alarm candidate is generated by; the instantaneous pan and tilt of the pan-tilt mechanism (23) angle information, fixed station starting coordinates (103), orientation reference data and numerical 21 By processing the terrain model (DEM) database (107) in an integrated manner, the fire focal point is three The dimensional calculated target location coordinates (104) are estimated. In the said positioning process, the digital positioning is performed by the local artificial intelligence processing unit (10). A ray-casting algorithm on the terrain model (DEM) database (107) It is being operated. The starting coordinates of the fixed station (103) are taken as the reference center, 5 Topographic surface in the direction corresponding to the orientation angle of the pan-tilt mechanism (23). A virtual ray vector is launched along the surface. This launched ray vector's topographical path... By identifying the point where the model intersects, the distance of that intersection point from the sensor center is determined. The geographical radial distance is calculated. The aforementioned ray tracing and intersection analysis is constant. By operating the surveillance stations during the initial setup phase as well; the formation of blind spots, 10 unnecessary sky scans and pointless analysis of atmospheric layers processes This is prevented at the initial stage. This calculated radial distance, beforehand The defined maximum scannable area outer limit (100) circular scan radii (101) It is checked whether it remains within the defined scanning volume or not. The radial distance is outside the defined scanning volume. If it remains, the relevant region is automatically eliminated from the system scan plan; suitable 15 If it remains within the range, the two-dimensional alarm coordinate (105) (X, Y) and three-dimensional geographic including the Z parameter of the calculated target location coordinates (104) Location estimation is performed. This generated geographic coordinate data is used only. It does not rely on geometric ray tracing calculations; it uses a high-resolution optical camera. (21) and is also supported by instantaneous anomaly pixels from the thermal sensor (22). The same 20 Multiple clusters of optical smoke or thermal pixels within the codomain scanning sector (102) When detected, the local AI processing unit (10) geostrategs this pixel location data. By superimposing the matrices, the accuracy of coordinate estimation can be improved. It optimizes. The geographical location and coordinate information in question is an absolute GPS. It does not have to be precise; the digital map interfaces of the intervention teams 25 they can use directly on the area, narrowing the incident zone and allowing teams to directly access the fire. It is a dynamic approximate coordinate estimation that directs to its focus. Thanks to this holistic structure... Operational reaction time is minimized and the first response is delivered at the most accurate point. Guidance is provided. 14. Converting Operator Feedback into Mathematical Adaptive Learning Fixed station operator interface (30) or central monitoring, recording and control by the operator Approval or rejection feedback provided via the interface (90) is only for instant alarms. It is not used for the purpose of terminating the situation. These operator decisions; The alarms are classified as verified alarms, temporary false alarms, and permanent false alarms; the alarm is 35. 22 co-field scanning sector where it occurred (102), alarm start timestamp (32), External atmospheric data (83) and anomalies collected by the environmental sensor unit (80) The classification type is analyzed relationally by the local artificial intelligence processing unit (10). Geographic data on persistent false alarm sources specific to the local station area is being collected. The matrices are processed primarily in the local AI processing unit (10) and then 5 transferred to the central monitoring, control and recording server (70) and into the relational database is indexed. The central monitoring, control and logging server (70) relationally indexes this transmitted data. The database identifies patterns that generate false alarms and common signs of real fires. to analyze characteristics and environmental behaviors specific to certain sectors Central monitoring, control and recording server (70) and local artificial intelligence processing unit 10 (10) uses operator feedback and associated event context set to deepen AI. It dynamically updates the learning models and weighting matrices. The adaptive model update process follows the general mathematical modeling below (Equation 5). It is carried out in accordance with: 𝑀𝑜𝑑𝑒𝑙 = 𝑀𝑜𝑑𝑒𝑙 + 𝛼 ⋅ 𝐹𝑒𝑒𝑑𝑏𝑎𝑐𝑘 + 𝛽 ⋅ 𝐶𝑜𝑛𝑡𝑒𝑥𝑡 15 Here, 𝑀𝑜𝑑𝑒𝑙 is the updated AI model parameter for the next period. its matrix, 𝑀𝑜𝑑𝑒𝑙 the currently executed model parameters, 𝐹𝑒𝑒𝑑𝑏𝑎𝑐𝑘 the parameters entered by the operator Structured alarm confirmation / rejection data, 𝐶𝑜𝑛𝑡𝑒𝑥𝑡 includes the geographical, temporal, and event location of the incident. The context matrix contains meteorological parameters, while 𝛼 and 𝛽 represent the local adaptation speed of the system. It refers to the predefined learning coefficients that determine. This mathematical 20 through the adaptation cycle, static models trained in a laboratory environment can be adapted to the field. The technical problem of blindness experienced under these conditions is eliminated. The operator's connection to the system... Each feedback loop it incorporates directly shapes the local weighting matrices of the AI ​​model. It functions as an active educational element that changes things. Confirmed data is centralized at specific intervals. processed by federated learning algorithms on the monitoring, control and recording server (70) 25 updated new model weights as data communication network (40) all local on the network Artificial intelligence is distributed to processing units (10). Thanks to this distributed learning approach, Uncontrolled false alarms that fully autonomous systems might generate are prevented, and legal notifications are made. responsibility criteria are met and a hybrid approach that incorporates operator experience into the learning process. A decision-making mechanism is being created. 30 15. The Impact of Meteorological and Environmental Data on Dynamic Scanning Strategy Meteorological data collected via the environmental sensor unit (80) are only used to generate alarms. not used in the verification phase of candidates, the pan-tilt mechanism (23) screening It also plays an active role in the dynamic determination of the strategy. Local artificial intelligence processing 35 23 Unit (10) processes wind direction and wind speed data to detect smoke in the event of a possible fire. and calculates the direction of frontal advancement as a vector and uses the resulting forecast data as a constant. The station provides information to the operator via the operator interface (30). Calculated spread In line with the vectors, those at risk or those previously deemed critical by the operator Scanning frequency and optical resolution of the defined codomain scanning sectors (102) 5 The parameters are automatically increased. This proactive approach ensures the system is not only available It's not about detecting anomalies, but about providing probability-based and preventative meteorological forecasting. It is carrying out the development of its mechanism. The same pixelated smoke or gas formation varies depending on microclimatic and seasonal conditions. They are classified by the system at completely different risk levels. Rainy or relative 10 A limited smoke component detected during high humidity winter periods indicates a low fire risk. It is classified as an anomaly carrying; high temperature, low humidity and dry summer Smoke formation with similar pixel density in different periods is a high-level risk indicator. It is considered as such. The system matches weather and environmental data with real-time visualizations. This In this way, it automatically generates the prioritization matrix without any human intervention. 15 Deep learning models operating within the local AI processing unit (10) provide instantaneous perception. by multiplying visual anomalies by the real-time "context set" parameters of the relevant scanning sector The system operates using real-time visual inputs and the context set detailed in Table 1 below. It generates an alarm candidate if the conditions correlate: Table 1: Alarm Candidate Context Set Parameter Matrix 20 Parameter Component Technical Description and Scope Sector (102) Identity of the co-field physical scanning sector where the alarm occurred Detection Type: Optical smoke, flame, thermal anomaly, or a combination / hybrid of these. variations Detection Time: The temporal continuity, persistence, and recurrence period of the anomaly. Meteorological Conditions via Environmental sensor unit (meteorology and soil) (80) incoming instantaneous temperature, humidity, pressure and wind matrix Operator Feedback: Errors previously identified by the operator for that sector. alarm suppression rules In this preventive early warning layer, local atmospheric and specific geographical conditions for a particular region are considered. soil characteristic data according to predefined critical threshold values ​​in the system A dynamic composite risk score (Rs) is obtained by weighting the normalized risk score. 24 is calculated. The aforementioned combined risk score (Rs) is derived from the following relational functional equation: It is updated in real time according to (Equation 6): 𝑅 = 𝑔 𝑃 , 𝐹 , 𝐻, 𝑈, 𝑇, 𝐷 Here, (Rs) represents the combined risk score of the related co-domain screening sector (102), (Ps) represents the basic risk score of the sector. Dynamic priority score, (Foperator) interface 5 by the operator in past periods. (H) central monitoring, control of the structured approval / rejection feedback coefficient entered via and the matrix of past alarm and fire records pulled from the database of the recording server (70), (U) The intensity value of moving people, vehicles, and objects detected in the scanned area. Instantaneous meteorological and soil data obtained from (T) environmental sensor unit (80) data. matrix and (D) is secondary physical verification from autonomous aerial vehicle (drone) (50) 10 It represents the data. Solution and optimization of the aforementioned functional equation (Rs); methods with constant coefficients, using adaptive coefficient methods or learning-based artificial intelligence methods This is being done. The calculated risk index directly generates a unique fire alarm. not for the purpose of; but to identify proactive environmental conditions that increase the likelihood of fire occurrence. 15 It is used for this purpose. The calculated combined risk score (Rs) exceeds predefined critical threshold values. In this case, a preventive measure, a warning, and independent of direct alarm generation. An early intervention mechanism is triggered. This warning layer triggers the relevant peer-to-peer scanning. The scan period frequency of the pan-tilt-zoom (PTZ) camera setup (20) for sectors (102) is 20 enhances the operator's situational response by automatically activating live video streams. proactively raising awareness and increasing the preparedness level of intervention teams It initiates preventive processes aimed at increasing the composite risk score (Rs) above critical thresholds. If it falls below this level, the scanning frequency is reduced to save hardware resources. This is provided. These long-term microclimatic and soil characteristics, collected continuously, are 25 data (86); drought trends of forest ecosystems, vegetation health, fire planning of post-ecological improvement / restoration processes and sustainable long-term Central monitoring, control and record keeping for the purpose of analyzing periodic forest management strategies. It produces data analytics outputs by archiving them relationally on the server (70). Figure 13 This shows the algorithm, and the numbers inside the circle represent algorithm step 30. These are the numbers. 16. Spatial Deduplication and Learning Process of Mobile Application Unit Data Contribution Multiple pixels from different mobile operators in the field via the mobile application unit (60) and geographic data, operating within a central monitoring, control and logging server (70) subject to coordinate estimation and spatial event deduplication algorithm. is held. The algorithm in question is simultaneously applied to different mobile application units (60). Data sent within a predefined time tolerance window (Delta t) of 5 and within the spatial radius tolerance matrix (Delta s) in the geographic information system (GIS) layer By intersecting them, it unifies duplicate reports in the field and creates a single mobile target alert point. It reduces to coordinate (61). Transferred to the system by the mobile application unit (60) and The structured dataset subjected to data correlation includes the following parameters:  Close-up anomaly image and video data of the target, 10  Mobile device location and altitude data of the sending device (62),  Compass and gyroscope-based orientation within the scope of mobile device orientation angles (pan-tilt) (63) angles (pan-tilt parameters),  Instantaneous atmospheric temperature, relative temperature, obtained via the mobile device atmospheric data sensor (64) Humidity and atmospheric pressure data, 15  Timestamp information based on the milliseconds in which the event occurred. Real-time multimedia data, centralized and enriched with these parameters, Pixel analysis is performed by deep learning models on the server, and the station is fixed. Operator interface (30) and central monitoring, recording and control interface (90) on the operator 20 It is submitted for approval with the status of "mobile-based alarm candidate". Thanks to this structure, in the field... actors can use the mobile application directly without needing a central operator connection. It can perform pinpoint incident reporting. Obtained from the mobile application unit (60) Positioned multimedia data provides only instant verification, showing the same event from different angles. 25 in the processes of observing and more precisely determining the location of the event is not used. The data in question includes all temporal data generated by the relevant alarm candidate, centrally sealed relationally with microclimatic and topographic context set parameters Monitoring, control and recording server (70) is archived in the database. Operator interface via fire verification selection area (33) or false alarm selection area (34) The final decision made is matched with this mobile crowdsourced dataset. These 30 labeled datasets are obtained. data hash, distributed and hybrid learning architecture within a centralized artificial intelligence model By incorporating their weights into the field learning process, similar events can be more accurately achieved in the future. It enables classification. 17. Secondary Physical Verification Method with Autonomous Aerial Vehicle (Drone) Integration 26 The rugged topography of the land, and the line of sight created by geographical obstacles. Due to (sight) limitations or microclimatic smoke / haze anomalies, some alarms are generated. candidates pixel level data from fixed station sensors or mobile application unit (60) This cannot be definitively confirmed. Operational in these complex terrain conditions mentioned. 5 to prevent blindness and to ensure legal reporting responsibility criteria The system dynamically activates the autonomous aerial vehicle (drone) (50) as a secondary physical confirmation layer. it is putting into operation. The autonomous task routing and verification mechanism operating within the system is sequential. It performs the following algorithmic steps:  Potential target alert via fixed station or mobile crowdsourced data 10 Calculation of coordinates,  Three-dimensional safe flight in accordance with digital terrain model (DEM) and environmental constraints. optimizing the route,  Fixed station operator interface (30) or central monitoring, recording and control interface (90) Receiving operator verification confirmation or automatic trigger signal via, 15  Autonomous aerial vehicle (drone) (50) on which the autonomous aerial vehicle is deployed starting point Automatic redirection, dispatch and data from location (54) to target region collecting,  After the task is completed and the operator seals the final decision, the autonomous aerial vehicle (drone) (50) Autonomous aerial vehicle automatically returns to its starting position (54). 20 Autonomous aerial vehicle (drone) (50) from autonomous aerial vehicle starting position (54) to target In order to perform a safe flight toward the alarm point (55), the digital terrain model (DEM) three-dimensional route optimization algorithm working on cross-section (53) data; static terrain elevation information, external geographic obstacle maps, wind direction / speed vectors, a total of 25 It simultaneously simulates flight distance and instantaneous battery consumption limits. The algorithm analyzes topographical obstacles and uses an autonomous aerial vehicle (drone) (50) to reach the target point. calculate the shortest straight-line distance vector between them (56) and this linear vector vertically on the predefined safe flight altitude threshold (offset) (52) By integrating it, it automatically creates the threshold autonomous flight route (51). The 30 in question When the route (51) is loaded via the data communication network (40) the autonomous aerial vehicle (drone) (50) It takes off automatically and is guided to the target coordinates. Autonomous aerial vehicle When the (drone) (50) reaches the target alarm point (55), its optical and thermal sensors close-up high-resolution spectral images and radiometric layers It collects temperature data. The collected real-time data is used by an autonomous aerial vehicle (drone) (50) 35 27 fixed station via data communication units, providing live or packet data. Live thermal monitoring with live optical monitoring screen (131) within the operator interface (30) By simultaneously transmitting the situational awareness of the operator to the screen (132), is being removed. 18. Recovering Autonomous Aerial Vehicle (Drone) Data for Artificial Intelligence Models 5 Close-up multispectral and radiometric data obtained from an autonomous aerial vehicle (drone) (50), It is not used solely for the purpose of verifying the currently active alarm candidate. From the field. This collected close-up anomaly data includes the topographic structure of the coordinates where the event occurred, and the instantaneous... microclimatic meteorological parameters and the operator's approval / rejection via the interface. Their decisions are relationally tagged and sent to a central monitoring, control, and logging server. 10 (70) and is transferred to the local AI processing unit (10) as a deep learning input dataset. Thanks to this algorithmic recovery process, the system model;  Complex geographic images that are pixelated or inadequate from the perspective of fixed cameras learning to distinguish anomalies from close-up data patterns,  By continuously updating deep learning datasets, it can identify false alarm patterns in the field. (cloud shadows, seasonal fog, dust hazes, etc.) are not precisely determining,  By developing local perception models specific to the geographical region and local topography future alarm candidate production strategy and dynamic risk management that it will implement in similar contexts. optimizing the priority score (Ps),  Taking into account past verification records via the local AI processing unit (10) 20 By doing so, we can improve the sensitivity of alarm generation and autonomous weather systems in similar environmental conditions in the future. vehicle (drone) (50) task delegation mechanism battery / range It maximizes its efficiency. 19. Unauthorized Moving Object Activities as a Preventive Fire Risk Indicator 25 Analysis A significant proportion of fire incidents occurring in open areas and forested regions, It is directly related to anthropogenic (human-caused) activities. The sparseness of natural forest cover. open areas, fire safety strips, roads on the site, or bodies of water such as rivers and lakes Unauthorized human presence and uncontrolled vehicle movement around the water sources constitute a potential 30 They are critical proactive early warning indicators in the emergence of fire risk. The invention in question... In this method, mobile object activities involving human and vehicle types are only considered in a conventional way. Not as a security input; but as the main component of a contextual and dynamic fire risk analysis mechanism. It is processed as one of the parameters. Figure 11 shows the functional flowchart. In the detailed process; local artificial intelligence processing unit (10), pan-tilt-zoom (PTZ) camera 35 28 object detection, classification and tracking on snapshots taken from the setup (20) It runs the algorithms. The local AI processing unit (10) performs the pixel detection in question. when performing this, the class type of the moving object (person / vehicle), the instantaneous velocity vector, and the progression direction and temporal dwell time within the related co-domain scanning sector (102) It performs contextual risk analysis by analyzing the data. The obtained object tracking data; 5 Identity of the co-field scanning sector (102) where the incident occurred, alarm start timestamp (32) and correlation with instantaneous measurement data (84) received from the environmental sensor unit (80). It is subjected to analysis. For example, microclimatically high temperature and critical level During a period of low relative humidity, people and vehicles would normally be unable to travel. 10 in a passive screening sector where no activity is expected, vehicle or human movement detection, top-level proactive risk by local artificial intelligence processing unit (10) It is rated as an indicator. 20. The Role of Moving Object Tracking in Scanning Strategy and Dynamic Risk Score Integration Detected unauthorized human or vehicle movements are directly recorded as a 15 by the system. These activities do not trigger a fire alarm. Instead, the aforementioned activities... contextual weighting coefficients that increase the dynamic risk priority level of the relevant geographic sector It is included in the decision-making mechanism. The local artificial intelligence processing unit (10) is involved in this type of pan-tilt in co-field scanning sectors where moving object activities are detected (102) The mechanism (23) automatically increases the scanning frequency of deep learning 20 raising the anomaly detection sensitivity thresholds of the models to more sensitive levels and Automatically activate live view mode in the operator interface when needed. This method allows the system to physically react to a fire. By monitoring high-risk areas at shorter intervals before starting, reactive measures can be implemented. It represents a transition from proactive approaches to an early warning architecture. Furthermore, this object is 25. Monitoring capability; preventing unauthorized forestry activities such as illegal logging, military / industrial environmental security applications and wildlife population dynamics It also simultaneously offers additional operational benefits such as monitoring. In the system that is the subject of the invention, static models trained in a laboratory environment are applied in the field. In order to overcome the inadequacies in their conditions, instantaneous 30 for each co-domain scanning sector (102) A continuous dynamic is achieved by normalizing meteorological parameters as they approach critical limits. The risk priority score (Ps) is calculated. This dynamic risk priority score (Ps) is... Local AI processing via the following weighted mathematical formulation (Equation 7) The unit (10) is updated instantly: 𝑃 = 𝑤 𝑇 + 𝑤 𝐻 + 𝑤 𝑈 + 𝑤 𝑂 + 𝑤 𝐷 35 29 Here, (Ps) is the instantaneous dynamic risk priority score of the targeted co-domain scanning sector (102), Instantaneous microclimatic and soil parameters obtained from the data of the (T) environmental sensor unit (80). matrix, (H) historical fire records and region-specific risk history retrieved from central server. data of unauthorized human or vehicle movements detected instantly in the (U) scanning area density and dwell time value, (O) instant scan / prioritization entered by the operator 5 request coefficient, secondary physical verification from (D) autonomous aerial vehicle (drone) (50) The outputs, parameters (wT, wH, wU, wO, wD) are determined by the location where the fixed monitoring station was established. pre-optimized according to the vegetation and topographic characteristics of the geographical region These represent the weighting coefficients. The higher the calculated dynamic risk priority score (Ps), the better. If this occurs, the scanning frequency and verification sensitivity of the relevant sector will be increased to a constant 10. Live video streams are proactively activated on the station operator interface (30). If the score is low, the screening period is extended and local artificial intelligence is used. The processor load on the CPU / GPU hardware resources of the intelligence processing unit (10) is reduced and By saving motor power of the pan-tilt mechanism (23) and the amount of mechanical movement Energy consumption is being optimized. 15 21. Final Alarm Decision Mechanism and Functional Model The system eliminates false alarms and improves operator operational awareness. In order to optimize fatigue, the generated preliminary "alarm candidates" are positioned at the operator's stationary position. 20 on the station operator interface (30) or central monitoring, recording and control interface (90) Following final approval, it is converted to a confirmed alarm status. This final approval... The decision-making mechanism is based on the following multivariate functional modeling (Equation 8) It is structured accordingly: Alarm Final = ℎ Alarm Candidate, 𝑅 , 𝐹 , 𝐷, 𝐷𝐸𝑀 25 Here, the Alarm Final is generated after operator approval and dispatched to external emergency response systems. The final confirmed alarm output is processed by the Alarm Candidate local artificial intelligence processing unit (10). the generated preliminary anomaly notification, 𝑅 the instantaneous composite risk score for the relevant geographic area, 𝐹 provided by the operator via the fire verification selection field (33) on the interface final approval / rejection feedback coefficient, (D) near 30 received from autonomous aerial vehicle (drone) (50). plan validation data and (DEM) based on the digital terrain model (DEM) database (107) These represent the pixel screen segmentation and masking parameters that are implemented. This functional structure controls alarm generation processes under contextual and geographical filters. This maximizes the autonomous stability and reliability level of the system, preventing false alarms. It minimizes the rate. In order to make the proactive structure of the system sustainable, the central 35 Monitoring, control and logging server (70), long term monitoring, control and logging of all this data accumulated in the field. By performing big data analysis, we can identify which environmental conditions pose a higher risk for specific sectors. It continuously learns what it carries through machine learning algorithms. This is centralized learning. The results and updated model weights are periodically sent to local AI processing units (10). By transmitting the data, the field parameters are dynamically updated. Thus, 5 The system will cease to be a static structure that only perceives current events and will instead encompass the future. A proactive early warning system that can anticipate potential risk situations and is self-renewing. It acquires the characteristics of a system. 22. Long-Term Distributed Learning Cycle and Federated Governance on a Central Server 10 Central monitoring, whose functional operating algorithm and flowchart are detailed in Figure 12, control and recording server (70); all networks connected to the system via data communication network (40) the main elements that constitute the long-term memory, archival layer, and collective intelligence of its architecture It is a component. Alarm candidates transmitted from fixed monitoring stations are interfaced with operators. approval / rejection labels provided via the mobile application unit (60) 15 close-up multispectral data collected from geographic information and autonomous aerial vehicles (drones) (50) These are brought together at the central layer and indexed in a relational database. Central basic operational, algorithmic and managerial aspects of the monitoring, control and recording server (70) Its functions include: (i) distributed learning coordination, (ii) data analysis and model updating, (iii) multi-functionality. (iv) long-term macro-pattern analysis and (v) 20 This includes the processes of archiving verification data and raw images. Within this scope... The server's basic functions consist of the following steps:  Distributed Learning and Federated Coordination: Local AI processing in the field. local deep learning models, anomaly detection patterns and from units (10) It analyzes weight matrices. 25 with operator experience in different geographical stations. local anomaly experiences acquired are federated in a controlled and context-sensitive manner. integrating multi-station data into a common master model within the learning architecture. It integrates them into a single learning pool. The new model is an updated version of the common core model. Reconfigure hyperparameter and weight matrices via data communication network (40) It distributes to all local AI processing units (10) in the field. 30  Long-Term Data Analytics and Model Optimization: Alarm accuracy in the field. their ratios, common characteristics of real fire indicators, and their dependence on environmental conditions. detection behaviors, pixel patterns that generate false alarms, and the hardware of sensors. It subjects the performance characteristics of operators to statistical and relational analysis. permanent false alarm sources (factory 35) marked via the false alarm selection area (34) 31 globally by labeling (chimneys, glowing surfaces, etc.) with a location and time stamp. It is updating the list of alarm suppression rules. This deep learning and performed on the central monitoring, control and recording server (70) Through statistical analysis processes, visual anomalies such as smoke and flames are naturally occurring. AI data that enables it to be distinguished from atmospheric fog, haze, or cloud formations. The collection is constantly being enriched. Collected from the field in unmarked or raw form. video data is securely stored on the central monitoring, control and recording server (70) storage; performing retrospective event analyses, artificial intelligence dataset. development processes, laboratory testing of new deep learning models, legal 10 and provides a technical advantage in terms of operating official operational record-keeping processes. The subject is a centralized management architecture, confirmed (Alarm Final) geographic information finalized with operator approval. The data includes: event geographic coordinates, timestamp, event class type, and wind propagation vector. along with data and environmental risk information; external geographic information systems, emergency response and disaster digital interfaces of management systems (AFAD, etc.), early warning monitoring systems of public institutions 15 to their networks automatically and in real time via industrial data communication protocols It has an external system integration infrastructure capable of transferring data. 23. Network and Resource Aspects of Hybrid, Distributed, and Centralized Decision / Learning Architectures Optimization In the system that is the subject of the invention, artificial intelligence learning and data processing processes are carried out using only monolithic 20 It is not run on a central server layer or solely on independent local units; a hybrid system where distributed and centralized learning mechanisms work synergistically together It is implemented in architecture. Local artificial intelligence processing units (10) are located on the fixed Edge computing on the micro-environmental and topographic conditions of surveillance stations alarm generation and pre-warning in real time by adapting with computing methods. While carrying out the evaluation processes, the central monitoring, control and recording server (70) is the network federated learning of long-term, macro-scale and multi-station anomaly patterns across the globe It learns through its algorithms. This two-stage learning hierarchy is applied throughout the system. maximizing local geographical adaptability as well as system-wide decision-making. This structure ensures the system's consistency. Thanks to this structure, the system consists of 30 completely different components. heterogeneous geographical regions with varying vegetation, topography, and climate characteristics Even when installed, each function requires no external manual software calibration. It automatically adapts to the specific local conditions of the region. In traditional architectures that operate entirely in the cloud or are centralized server-based, high Due to the necessity of seamlessly transmitting multiple high-resolution video streams to the center, a serious 35 32 Data transmission delays (latency) and network bandwidth bottlenecks occur. This invention addresses the distributed and centralized division of labor architecture, which involves heavy processing load. This requires computer vision, thermal anomaly analysis, and the direct generation of preliminary "alarm candidates". Data is performed within the local AI processing unit (10) at the field boundary. uninterrupted 5 to the central monitoring, control and recording server (70) via the communication network (40). Instead of raw video streams, only qualitative metadata of locally generated alarm candidates is used. instantaneous measurement data (84) and operator approved / rejected relational structured feedback Tags are being transmitted. The aforementioned distributed and centralized server combination relates to the network and operations. In general, it offers the following technical advantages:  By eliminating the need for raw image transmission, the data communication network (40) 10 The instantaneous bandwidth consumption requirement is statistically reduced and optimized.  Local anomaly detection and alarm system that operates independently of network congestion. The candidate's production processes minimize latency,  The same from different fixed surveillance stations or mobile application units (60) Simultaneously transmitted multi-source data for geographical targeting, centralized monitoring, control and recording 15 subjected to spatial correlation and deduplication algorithm on server (70) combined under a single operational event / record,  Dependence of sensing processes on a single sensor output or a single piece of hardware by blocking and through optical, thermal, meteorological and crowdsourced multiple data fusion, a single Sensor dependency is reduced and final event accuracy is increased, 20  Intervention by minimizing the positioning error margin of the target alarm point (55) The response time of the teams, their arrival at the scene, and their response time to the fire are shortened. between the local station and the central monitoring, control and recording server (70) The distinction between modular and scalable architecture allows the system network to be established at different scales. and allows for flexible regional expansion. Included in the system each new fixed monitoring station or external environmental sensor unit (80) to be installed, the existing topological structure on the data communication network (40) and central monitoring, control and recording The server (70) can be modularly operated using plug-and-play technology without compromising its stability. 30 that can be integrated into distributed network architectures and adapted to heterogeneous environmental conditions It offers an industrial infrastructure.

Claims

33 REQUESTS 1. Wide area surveillance, early incident detection, risk assessment, and adaptive alerts. It is a system for verification, and its feature is; - a high-resolution optical camera positioned on a power supply pole (12) (20) pan-tilt-zoom (PTZ) camera setup (21) housing thermal sensor (22), pan-5 including the tilt mechanism (23), reference sensors (24) and data communication unit (11) at least one fixed surveillance station, - fixed station starting coordinates (103) of the fixed observation station, reference position, orientation and slope data received from sensors (24) and Digital Terrain Model (DEM) Maximum scannable area outer limit (100) by processing database (107) data together 10 radially along the circular scanning radii (101) of the surveillance area within it scanning bands and these bands with varying angular widths on the Earth's surface equal area scanning sectors covering physically approximately equal areas (102) separating; pan-tilt mechanism (23) for each co-domain scanning sector (102) pan and tilt Zoom 15 of the pan-tilt-zoom (PTZ) camera setup (20) with its parameters passive excluded from scanning by line of sight vector (106) analysis which calculates its parameters. screen areas (122), masked land area screen areas (128) and maximum land The out-of-scan area whose height remains above the vertical atmospheric threshold area (Elevation offset) (129) removing the abandoned atmospheric region screen area (123) from the scan plan; each equal area Instantaneous environmental data, past alarm and fire records for the scanning sector (102), 20 using operator feedback and unauthorized human or vehicle (41) activities dynamic calculating the risk priority score (Ps) and determining the screening frequency and optical frequency based on that score. The resolution is determined by; high-resolution optical camera (21) and thermal sensor (22) By processing the data using edge computing methods, it detects fire signs and identifies people, vehicles, and wildlife. It detects animal activity; it converts detected anomalies into alarm candidates instead of definite alarms. 25 and live optical monitoring screen (131) and live thermal monitoring screen (132) only when an alarm candidate is generated or a manual review is requested by the operator. a local AI processing unit that activates when (10), - generated alarm candidates are displayed on the fixed station operator interface (30) and / or central monitoring, The registration and control interface (90) presents the operator with fire verification 30 Confirmation / rejection feedback configured via selection area (33) and false alarm selection area (34) the interfaces that enable it to provide the notification, - a fixed surveillance station, the interfaces in question, and at least one central monitoring and control system. and the data communication network (40) connecting the registry server (70), - operator feedback to the codomain scanning sector (102) to which the alarm belongs, alarm 35 start timestamp (32), external atmospheric data (83) and anomaly type 34 artificial intelligence models that record data relationally by labeling; based on labeled data. updating their weights and decision mechanism filters through federated learning, and updated model weights via data communication network (40) local artificial intelligence processing It includes the central monitoring, control and recording server (70) which distributes to the unit (10).

2. The system compliant with Claim 1, its feature is; high resolution optical camera (21) and thermal Computer vision of raw images from sensor (22), thermal anomaly analysis and alarm For the production of the candidate, the local operating and uninterrupted raw data communication network (40) Instead of video, metadata of alarm candidates, real-time measurement data (84) and operator approval / rejection With the local AI processing unit (10) that transmits the tags, the data in question is relationally processed. Central monitoring that records and performs global model hyperparameter optimization, It includes the control and registration server (70).

3. The system complies with Claim 1 and its features include a high-resolution optical camera (21) and thermal Data obtained from the sensor (22), mobile application unit (60) mobile device location and 15 altitude data (62), mobile device orientation angles (Pan-tilt) (63), mobile device atmospheric data sensor (64) data and positioned multimedia data contextually together by evaluating and creating a candidate for alarm, blind spots caused by topographical obstacles It includes a local AI processing unit (10) that enables verification.

4. The system complies with Claim 3 and its features include a high-resolution optical camera (21), thermal Cross-reference the sensing results obtained via the sensor (22) and the mobile application unit (60) which subjects the results to verification, contextually weights those results, and pre-determines by suppressing pixel data that generates false alarms under specified microclimatic threshold conditions It includes a local AI processing unit (10) that dynamically reduces the false alarm rate. 25 5. The system complies with Claim 1 and its features include: fixed surveillance stations and mobile application units. (60) and / or via autonomous aerial vehicles (Drone) (50) relating to the same geographical target By subjecting the received duplicate alarm records to temporal and spatial correlation, a single event can be identified. It includes a central monitoring, control and recording server (70) that combines the records under one name. 30 6. A system that complies with Claim 1, and whose characteristic is that it is connected to or integrated with a fixed surveillance station, solar and atmospheric radiation sensors (81) and soil moisture and characteristic sensors (82) containing instantaneous measurement data (84), meteorological critical values ​​(85) and soil characteristics 35 environmental sensor unit (Meteorology) which transmits data (86) to local artificial intelligence processing unit (10) (80) is that it contains soil).

7. A system that complies with Claim 6, and whose characteristic is; a combined risk score (Rs) and / or dynamic risk If the priority score (Ps) exceeds the predefined critical threshold values, the relevant peer 5 increasing the scanning frequency of the pan-tilt mechanism (23) for the field scanning sectors (102), fixed generating visual and / or auditory alerts on the station operator interface (30) and / or autonomously Critical threshold decision logic that generates commands for dispatching the aircraft (Drone) (50) It contains the layer (87).

8. Automated scanning based on geographic terrain awareness in a wide-area surveillance system, It is an alarm candidate generation and adaptive verification method, the feature of which is; - fixed station starting coordinates (103), from reference sensors (Position, direction and slope) (24) orientation data received and Digital Terrain Model (DEM) database (107) data collection, 15 - Circular scanning of the surveillance area within the outer limit of the maximum scannable area (100) Separation into radial scanning bands along the radii (101) and each radial scanning within its band, it has varying angular widths, physically located on the Earth's surface. co-area scanning sectors covering areas of approximately equal size (102) creation, 20 - Using line of sight vector (106) and topographic surface intersection analyses, outside of scanning left passive screen areas (122), masked land segment screen areas (128) and the maximum terrain height above the vertical atmospheric threshold area (Elevation offset) (129) from the scan plan of the remaining excluded atmospheric region screen area (123) removal, 25 - Calculation of the dynamic risk priority score (Ps) for each co-domain screening sector (102) and the frequency of screening and optical screening according to the risk score of the co-field screening sectors (102) prioritizing in terms of resolution, - Vertical field of view angle (Tilt FOV) (108) for each co-field scanning sector (102), horizontal Field of view angle (Pan FOV) (109), pan and tilt orientation parameters and zoom 30 Calculation of parameters and pan-tilt-zoom (PTZ) camera setup (20) Scanning via pan-tilt mechanism (23) according to calculated parameters to be carried out, - Edge computing of high-resolution optical camera (21) and thermal sensor (22) data processing by method, fire signs and human, vehicle and wildlife activity 35 36 detection and identification of anomalies, and using them as potential alarms rather than definite alarms. creation, - alarm candidate fixed station operator interface (30) and / or central monitoring, recording and presented to the operator via the control interface (90) and configured from the operator Receiving approval / rejection feedback, 5 - operator feedback codomain scanning sector (102), alarm start timestamp (32), external atmospheric data (83) and anomaly type are labeled for central monitoring and control. and transmitted to the registration server (70) and updated using the labeled data by artificial intelligence model weights to local artificial intelligence processing units via data communication network (40) (10) includes the distribution process steps. 10 9. The method is compliant with Request 8 and its feature is; Digital Terrain Model (DEM) database (107) Target location calculated using fixed station starting coordinates (103) three-dimensional line-of-sight vectors (106) between coordinates (104) using ray tracing method The calculation involves the intersection of the relevant line-of-sight vectors with the topographic surface of the earth, 15 determining the maximum terrain elevation at geographical points that intersect the line of sight determination of the maximum terrain height at the determined height, vertical atmospheric threshold area (Elevation offset) (129) addition, the upper atmosphere remaining above the said threshold area and sky regions as the screen area of ​​the atmospheric region excluded from scanning (123) and physical Masked land area screen 20 showing inaccessible or fire-free regions The process includes removing area (128) from the scanning plan.

10. This method complies with Claim 8 and its characteristic is; - within the scope of implementing image processing and operator interaction, maximum 25 of the calibration images obtained within the outer limit of the scannable area (100) by combining the static elements, a 360-degree static background reference of the environment is created. creation of panoramic screen (120), - in summer / winter and night / day operating modes of the static panoramic display (120) visualization and as a static map layer instead of continuously updated live video. eclipse, 30  The horizontal and vertical orientation angles of the pan-tilt mechanism (23) panoramic screen horizontal- Matching pixel positions with vertical angle synchronization (138), instant scanned panoramic screen of the region with instant scanned band display (137) and alarm candidate its location is indicated by a visual alarm symbol (39),  The live optical monitoring screen (131) and the live thermal monitoring screen (132) are uninterrupted for 35 minutes. not being run and only the alarm candidate being generated in the co-domain scanning sector (102) or 37 operator's live thermal display target indicator (135) and / or live optical display target if he requests a manual inspection via indicator (136) The activation process includes the necessary steps.

11. The method is in accordance with claim 10, and its feature is; normalized 5 on a static panoramic screen (120). Instantaneous scan sector screen segments (121), belonging to co-field scan sectors (102) risk levels and two-dimensional alarm coordinates of the alarm candidate (105), different color codes, with a heat map layer consisting of graphical signals and / or pixel density variations Visualization involves including the process steps.

12. This method complies with Claim 8 and is characterized by the dynamic determination of scanning priorities. In this context, the dynamic risk priority score (Ps) for each co-domain screening sector (102), Instantaneous measurement data (84) received from the environmental sensor unit (Meteorology and soil) (80), central past alarm and fire records on the monitoring, control and recording server (70), operator past approval / rejection feedback and unauthorized persons or vehicles identified in the relevant sector (41) 15 Calculation using the intensity and duration of activities and high-risk co-locations. screening sectors (102) have shorter time intervals and / or compared to low-risk sectors Scanning pan-tilt mechanism (23) so that it can be scanned with higher optical resolution Prioritization of the cycle involves including process steps.

13. This method complies with Claim 8 and its characteristic is the performance of geographical coordinate estimation. Within this scope, the pan and tilt orientation angles of the pan-tilt mechanism (23), pan-tilt-zoom (PTZ) The zoom parameters of the camera setup (20) are derived from the reference sensors (Position, orientation and slope) (24) received orientation data, fixed station starting coordinates (103) and Digital A view of the topographic surface using the Terrain Model (DEM) database (107) data 25 The creation of the line vector (106) is the intersection of the said vector with the topographic surface. Determination of the point was obtained from the high-resolution optical camera (21) and thermal sensor (22). The correlation of the detected anomaly pixel clusters with the intersection point in question, calculated Determination of target location coordinates (104) and data communication network (40) This includes the steps involved in transmitting the information to system components and external emergency response interfaces. 30 14. This method complies with Claim 8 and its characteristic is; - Within the scope of implementing operator-interactive adaptive learning, local artificial intelligence Fixed station operator interface (30) of the alarm candidate generated by the processing unit (10) and / or presented to the operator via the central monitoring, recording and control interface (90) and 35 38 Alarm candidate verified alarm or false alarm with fire verification selection field (33) selection area (34) and its classification as temporary or permanent false alarm, - operator feedback alarm start timestamp (32), two-dimensional alarm coordinate (105), external atmospheric data (83), co-field scanning sector (102), anomaly type and geographic, Labeling with temporal, microclimatic and topographic context data, 5 - Transmission of labeled data to the central monitoring, control and recording server (70), AI model weights and decision mechanism via federated learning on the server updating the parameters and the updated model weights in the data communication network (40) distributed to local AI processing units (10) where each unit is located Adapting to regional conditions involves process steps. 10 15. This method complies with Claim 14 and its characteristic is the implementation of contextual alarm filtering. Within this scope, recurring false alarm sources are identified based on operator feedback. tagging with geographic location, timestamp, and event type information, specific geographic 15 coordinates associated with date ranges, days of the week and / or time zones during the day Establishing temporary or permanent alarm suppression rules, applying these rules to new alarms contextual evaluation of candidates by local artificial intelligence processing units (10) the application as a filter and temporary alarm suppression rules at the end of the defined period It includes the steps for automatic termination.

16. This method complies with Claim 13 and is characterized by autonomous exploration and physical verification. As part of the implementation, the calculated target location coordinates of the alarm candidate Determining the target alarm point (55) according to (104), autonomous aerial vehicle starting position (56) is the shortest straight-line distance vector between (54) and the target alarm point (55). determination of the Digital Terrain Model (DEM) vertical cross-section (53) and safe flight altitude threshold 25 Calculation of the threshold autonomous flight route (51) using (Offset) (52), operator approval Target alarm via the data communication network (40) of the autonomous aerial vehicle (Drone) (50). being sent to point (55), close-up taken via autonomous aerial vehicle (Drone) (50) Inclusion of optical and thermal data as a secondary confirmation layer in the alarm verification process and When the verification process is completed, the autonomous aerial vehicle (Drone) (50) autonomous aerial vehicle 30 It includes the steps to return to the starting position (54).

Images